Score:6

RAID with different drive types possible?

in flag

In a high density storage server, which is in a RAID array, do all drives have to be of same type. Such as a mix of drives SSD, SATA? In a RAID array most controller require only that all drives be of the same capacity but what about drive type?

cn flag
Can not be answered. RAID is a concept - the implementation may differ. It is up to the controller and should be mentioned in the documentation (or, seriously, you may have to try it out - because it is rare to mix that in the real world it may well have been omitted). But there is nothing in RAID - as defined - that forbids it.
Sam avatar
ms flag
Sam
I actually did something like this with ZFS for a hobby search engine project at work, I got SSD-like speed for reads and ZFS transactions hid a lot of the random write penalty on the HDDs. I wouldn't recommend it in production though, if you can't add more RAM, L2ARC is a better option if you've got ***good, high-endurance*** drives - it's just that this seemed like a good way to get the read speed of RAID0 and the redundancy of RAID1 with the second-hand freebies I had on hand...
Score:14
ca flag

Hardware RAID controllers from LSI, DELL, HP etc. does not allow mixing disks with different interfaces (eg: SATA and SAS) in a single array. What you can do is to create two different arrays, each for a specific interface protocol - a SATA array and a SAS one, for example.

Software RAID does not share this limitation - basically any block device (even a loopback device) can be part of any arrays. However, mixing different disk technologies is generally discouraged to avoid an unbalanced array (performance wise). For cache drives, as ZFS L2ARC or LVM dm-cache, things are different - here you actually want a faster drive. So, for example, using an NVMe cache in front of a SATA array is perfectly fine.

br flag
Another useful configuration on Linux is a software RAID1 between an SSD and a RAID6 volume, with the `--write-mostly` flag applied to the RAID6, this gives you the performance of the SSD, but you still have very good fault tolerance.
shodanshok avatar
ca flag
@SimonRichter This requires an SSD as big as the RAID6 volume, which kind of defeat the purpose of a true SSD cache: to have a smaller SSD tier to accomodate the working set only.
br flag
I think of it the other way round: the SSD is the main drive, and the RAID6 volume is the backup in case the SSD dies.
sa flag
@SimonRichter then I am not sure of the point of the RAID6 - are you throwing a pile of old drives together? no matter how big your SSD is, an HDD of the same capacity is much cheaper, and if you are worried about concurrent failure (given the higher failure rates of SSDs) you can have an SSD with 2 HDD's in RAID-1
joshudson avatar
cn flag
@user253751: The cheapest density per byte isn't the biggest capacity, or he might have concatenated SSDs.
sa flag
@joshudson I believe usually the cost of the electricity and extra ports means that it is best to go with big drives.
Peter Cordes avatar
ke flag
@SimonRichter: Given the usual size differences between SSDs and HDDs, a write-mostly mirror for an SSD would probably only need part of one HDD. Or a RAID0 or RAID5 of partitions on multiple HDDs, the rest of which you use for their own thing. (Or if you're taking a small slice of a large number of drives to match the total size of your SSD, then sure maybe RAID6, but the more disks you have in a RAID6 or RAID5, the worse the scattered write performance is, right?)
br flag
Exactly -- I have lots of data in the RAID6 anyway, and I've set up a small volume to follow the SSD. The (battery backed) write cache makes the scattered write performance bearable, and I can hot-swap the SSD when it fails.
Score:5
kz flag

Hardware RAID: No, you cannot. Except maybe mixing up SATA with SAS.

Software RAID: Yes, you can! But, you probably shouldn’t do that as combined RAID volume performance would be limited with the slowest drive in the resulting RAID set.

P.S. There’s an exception when faster drives act as a caching or log devices while slower do capacity. Think about ARC and ZLOG in case of ZFS and about Mirror-Accelerated-Parity in case of Storage Spaces and ReFS combo. You can Google these referenced acronyms to satisfy your curiosity.

I sit in a Tesla and translated this thread with Ai:

mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.